Decisions in a Box.

Back in the day of shrink wrap software there was a leader at Microsoft that characterized what we did as a software team was packaging the team's collective intelligence into a box. What we were selling was the brilliance of the best programmers we could hire and then trying to capture as much of intelligence focused on a product as we could on a CD. If anybody can remember who that was, I'd love to credit them.

What I was thinking was that in this age of LLMs and approaching AGI, where knowing things has become devalued, and knowing how to do things (at least things in the information domain) is becoming increasingly devalued, we are left with figuring out what exactly we should focus on capturing so that we can put it in a box and sell it. Or more likely, package up as a service and rent it.

Having spent some time now working with the cutting edge of LLM based coding assistants, I find that what I am increasingly spending the most time doing is adding the human context in which the software is intended to exist. And the act of making choices to translate that context feels a lot like aesthetics. Like blue is better than green for this element. Or maintaining a deployment document is going to make life much easier for anyone who has to deal with this at some future time.

I am putting as many choices as I can make in the box that ends up for sale. Crossing my fingers that it will remain valuable for a while still.